Pass the remaining engine context-length budget to model-owned vision processors through VisionPreprocessingContext. Preserve Nemotron batched engine fields and recognize llm_config as a text_config alias. Use the merged upstream llm-multimodal revision f0985ef65967615db2c79279aa07818499301bfd. Co-authored-by: Codex <noreply@openai.com> Signed-off-by: Bugen Zhao <i@bugenzhao.com>